Openai for the first time the new “Thinking” models and the coding agent, as it seeks to stay at the forefront of the artificial intelligence package

Photo of author

By [email protected]



Openai has Two models of artificial intelligence were released “Thinking” She says she is more able to do so, in addition to an open source AI agent that helps the computer programmers icon, as the company seeks to obtain the initiative for its competitors.

The open source coding agent, called Codex Cli, is the first time since 2019 that Openai has provided an important open source tool.

Other new models are the O3 wide-ranging version, which Openai says it is the most advanced artificial intelligence system, as well as a smaller model, but it is more efficient called O4-MINI.

“These are the first models in which senior scientists tell us that they are producing new and legally beneficial ideas,” Greg Brockman, President of Openai, said in the announcement of new products on Wednesday.

The models will be immediately available for Chatgpt Plus and Pro Pro users, as well as institutions that use their teams that focus on institutions and API products.

The version of the new models comes at a time when Openai faces pressure to show that it remains at the forefront of developing artificial intelligence. Earlier this year, Dibsic has raised the traditional wisdom about the American technological edge of artificial intelligence such as Openai for years. Deepseek’s R1 simulates the “Thought Series” offered by Opeenai’s O-Series models. The fact that Deepseek’s R1 was also an open model – which means that people can download it for free and customize it easily – Many institutions are tilted in favor of publishing such open source models. In contrast, most Openai’s models cannot be accessed on a paid basis through the API.

At the same time, Openai also faced increasing competition from other royal models. In February, AI HotHROPIC became the first company to provide a model that combines quick -likening answers with the ability to perform thinking “a series of thought” step by step if it requires it. The ability to determine when thinking is required and when the fastest answer will do is a trick that has not yet matching Openai. Then, last month, Google The Detection of the Gemini 2.5 Pro modelThinking model that won the Openai’s O3-MINI model on many criteria.

On Wednesday, Openai moved to try to restore the lead in thinking models. The company says from the O3 and O4-MINI models now in many different criteria-although none of these results have been independently verified. It also says that the models have the ability to use other software tools independently, such as web browsing and coding environments, without the need for a specific claim to do this by the user.

In a experimental presentation of the O3 capacity that was exposed to the live broadcast on Wednesday, artificial intelligence researchers showed an analysis of the physics research poster from 2015 and then independently searching for the web to find more modern research and comparison of the results. They also showed that it was independently decided to run the Bithon code to solve the various challenges of mathematics and coding.

Openai said that O3 and O4-MINI have the ability to directly think about visual information, such as graphics, graphics, or images-even those that may be blurry or poor quality. The company said the models also knew how to process images as part of the thinking process.

Meanwhile, the new Codex CLI coded agent is designed to run on the user device, and take advantage of the cloud-based connection to O3-MINI models from Openai to help him cause the reason, but then allow him to use other software tools that have been published locally. Codex Cli not only suggests lines of code, but can independently decide to use a variety of different tools to help them complete the task.

The company said that Codex Cli will soon be able to take advantage of the capabilities of the GPT-4.1 model earlier this week.

To encourage developers to experience Codex Cli, Openai said it has created a $ 1 million fund that would spend $ 25,000 in API credits for promising projects.

Openai said that O3 used about 10 times the computing capacity of training and it took it to create its O1 model, which is its best model.

This story was originally shown on Fortune.com



https://fortune.com/img-assets/wp-content/uploads/2025/04/GettyImages-2208831531-e1744822295568.jpg?resize=1200,600

Source link

Leave a Comment